Docs / Reference

Performance Numbers

Measured values, with their conditions. Speed depends on your CPU and your link.

Method

Every number below is a real measurement from a dated benchmark run. "Up to" values are the highest recorded. Encryption is always on; there is no unencrypted mode.

Transfers are CPU-bound on fast links (AES-256-GCM throughput) and wire-bound or latency-bound on slow ones. Expect lower numbers on older CPUs, and higher numbers on newer ones. WAN numbers are from the July 2026 tuning wave; a 17-hour soak on the same link completed 102 of 102 transfer probes with matching checksums and zero reconnects.

These figures supersede older rounded figures elsewhere on this site.

Throughput

Same machine (loopback, no wire limit)

ModeMeasuredConditionsLimiting factor
Direct transfer (no mount)up to ~660 MB/s3 GB file, May 2026AES-256-GCM CPU
Through the mount (FUSE)up to ~276 MB/s1 GB file; runs range 152-276 MB/s across benchmark roundsFUSE kernel overhead; varies with thermal and cache state

Intercontinental link (Singapore to Romania, ~200 ms RTT, July 2026)

Data-center to data-center, 1 Gbit/s ports, through the bridge, encryption on.

ScenarioMeasuredConditions
Bulk transfer33-37 MB/s sustained10 GiB, both directions, 5 trials each, zero stalls
Cold on-demand read through the mount20-30 MB/sSame pairing. sshfs on the identical path: 1.33 MB/s
Paced video playback through the cold mount0 stalls, worst read 70 ms1 GB file, 3 MB/s player, default read-ahead. Without read-ahead: 62 stalls

The bulk rate sits within about 26% of a raw unencrypted TCP stream on the same path. That gap is the encryption and framing cost. In one direction KeibiDrop ran above raw single-stream TCP, because it uses parallel streams.

Home link (500 Mbps cap, 24 ms RTT)

On a fast home line the wire is the limit, not KeibiDrop: direct transfers measured 56.9 MB/s up and 54.3 MB/s down on a 3 GB file, which fills the 500 Mbps line.

The bridge path measures about 9% slower than a LAN path for the same 1 GB transfer.

Sync latency

EventMeasuredConditions
Edit reaches the other peer0.56-0.61 sReal WAN, ~17 ms RTT, through the production bridge, each direction
Conflict file appears after the winning edit+0.02 sSame WAN setup
Cold read of a small file (512 KiB)30 msLoopback

Transfer granularity